matlab-profile-1.png
Launch matlab from the login node by first loading the module and then running matlab.
matlab-profile-2.png
The matlab GUI will launch and you can load our default cluster configuration.
matlab-profile-3.png
Select the Manage cluster profiles. Option from the environment section of the toolbar.
matlab-profile-4.png
This will open the cluster profile manager. You will want to import our profile.
matlab-profile-5.png
Import the profile from: /share/apps/samples/matlab-profile/nic_cluster_general.settings
matlab-profile-6.png
Next you will select the profile nic_cluster_general and set it as the default. It is configured for 12 nodes by default. You can use up to 160 by modifying the profile's numworkers range setting.
matlab-profile-7.png
Now you can use matlab as normal, once your matlab script is loaded, or written, you can launch the parallel job.

Matlab Cluster Profile

Mathworks has an excellent tutorial of how to use the cluster in the matlab scripts here.

This tutorial shows you how to configure the cluster profile correctly for our cluster.  We recommend using the tutorials above to write your matlab scripts.

To start a Cluster Profile job follow the instrutions for connecting to the cluster with X-forwarding.

Once you are sucessfully connected:

Load the module file for the version of matlab you wish to use.

[joemier@login-16-30 ~] $ module load matlab-2014

Start matlab: If the graphical interface fails to load make sure you enabled X-forwarding on your cluster session. See SSH/X-Win32 Tutorials.

[joeminer@login-16-30 ~]$ matlab

Once inside the gui on the home screen there is an environment section in the toolbar. Select the Parallel drop down, then select manage cluster profiles.

A new window will open and you will import the profile I have built using the import button in the toolbar.

Navigate to /share/apps/samples/matlab-profile and there is a nic_cluster_general profile there.

Once this profile is imported set it as the default, and run the validation on it.

It is currently set to run on 12 cpus and this can be adjusted in the cluster profile by changing the numworkers range setting. We have been able to validate successfully with 150 workers or less. Changing the value in this field will change how many nodes your job requests, and matlab should submit the jobs for you so to run your parallel code you could load your script in the gui and run it like you would locally.